This could have the unintended consequence of encouraging under-agers to ask more 'adult' questions in order to try to trick it into thinking they're an adult. Analogous to the city that wanted to get rid of rats, so offered a bounty for every dead rat, and to the surprise of nobody except policy makers, the city ended up with more rats, not less. (lesson: they thought they were incentivising less rats, but unintentionally incentivised more)

The padding in OpenAI's statement is easy to see through:

> The model looks at a combination of behavioral and account-level signals, including how long an account has existed, typical times of day when someone is active, usage patterns over time, and a user’s stated age.

(the only real signal here is 'usage patterns' - AKA the content of conversations - the other variables are obfuscation to soften the idea that OpenAI will be pouring over users' private conversations to figure out if they're over/under age.).

Worth also noting 'neutered' AI models tend to be less useful. Example: Older Stable Diffusion models were preferred over newer, neutered models: https://www.youtube.com/watch?v=oFtjKbXKqbg&t=1h16m